THE CONSEQUENCES OF CHANGING THE TOP PREDATOR IN A FOOD WEB: A COMPARATIVE EXPERIMENTAL APPROACH

Ecological Monographs, 68(1), 1998, pp. 1 23 1998 by the Ecological Society of America THE CONSEQUENCES OF CHANGING THE TOP PREDATOR IN A FOOD WEB: A COMPARATIVE EXPERIMENTAL APPROACH MARK A. MCPEEK Department of Biological Sciences, Dartmouth College, Hanover, New Hampshire 03755 USA Abstract. Changing the top predator in a food web often results in dramatic changes in species composition at lower trophic levels; many species are extirpated and replaced by new species in the presence of the new top predator. These shifts in species composition also often result in substantial alterations in the strengths of species interactions. However, some species appear to be little affected by these changes that cause species turnover at other positions in the food web. An example of such a difference in species responses is apparent in the distributions of coenagrionid damselflies (Odonata: Zygoptera) among permanent water bodies with and without fish as top predators. Enallagma species segregate between ponds and lakes that do and do not support fish populations, with each lake type having a characteristic Enallagma assemblage. In contrast, species of Ischnura, the sister genus to Enallagma, are common to both fish and fishless ponds and lakes. Previous research has shown that Enallagma species segregate because they are differentially vulnerable to the top predators in each lake type: dragonflies in fishless lakes and fish in fish lakes. This paper reports the results of a series of laboratory and field experiments quantifying the mortality and growth effects of interactions in the food webs surrounding Enallagma and Ischnura species in both lake types. These results are compared to determine how features of the food web change to force segregation of Enallagma species between the lake types but permit Ischnura species to inhabit both. The results of experiments conducted in a fishless lake show that damselflies are not food limited in this lake type, but that they do strongly compete via interference mechanisms. Interference effects between the genera are symmetrical. Ischnura species have substantially higher growth rates than Enallagma species under all conditions in fishless lakes. Although both Enallagma and Ischnura experience substantial mortality from predation by dragonflies (Anax and Aeshna species, the top predators in fishless lakes), these dragonflies display a significant bias towards feeding on Ischnura. Mortality rates due to dragonfly predation are not density dependent. The results of experiments done in a fish lake indicate that damselflies are food limited and thus compete for resources in fish lakes. Ischnura growth rates are also substantially higher than Enallagma species in the fish-lake system. Dragonfly species that coexist with fish (Basiaeschna and Epitheca species) do not impose significant mortality on coexisting damselflies, but they do compete for resources with the damselflies, and they may also generate feeding interference in the damselflies. Fish impose significantly higher mortality on Ischnura species than on coexisting Enallagma species, and this mortality is negatively density dependent. The coexistence of Enallagma and Ischnura species is fostered in both lake types by trade-offs in their abilities to avoid predators and to utilize resources. Native Enallagma species are better at avoiding coexisting predators in each lake type, but these abilities come at the expense of the ability to utilize resources effectively and to avoid the predator found in the other lake type. In contrast, Ischnura are better at utilizing resources in both lake types, but these abilities come at the expense of effectively avoiding both fish and dragonflies. Understanding the trade-offs faced by species at similar trophic positions within a food web is critical to predicting changes in food webs following major environmental perturbations such as changing the top predator. Key words: coexistence; community structure; density dependence; Enallagma; food limitation; food web; Ischnura; Odonata; predation; resource competition; trade-offs; trophic structure. INTRODUCTION Understanding how community structure is altered by dramatic environmental changes is a major focus in ecology. These efforts are motivated by comparisons Manuscript received 18 September 1996; revised and accepted 27 January 1997. 1 of natural communities existing at different points along environmental gradients that generate these dramatic changes (e.g., intertidal zonation patterns [Connell 1961, Paine 1966, 1974, Lubchenco 1978, 1980], lakes with different numbers of trophic levels [Brooks and Dodson 1965, Dodson 1970, 1974, Sprules 1972, Zaret 1980, Vanni 1986, 1988, McPeek 1989, 1990a,

ECOLOGICAL MONOGRAPHS Saturday Oct 10 01:01 PM Allen Press DTPro emon 68 103 Mp 2 File # 03sc 2 MARK A. McPEEK Ecological Monographs Vol. 68, No. 1 Persson et al. 1992, Arnott and Vanni 1993], and terrestrial and lake communities along nutrient gradients [Tilman 1982, 1988, McQueen et al. 1986, 1989]). Much of current theory explores the consequences of these dramatic ecological changes using models that assume that whole trophic levels respond as units (e.g., Fretwell 1977, Oksanen et al. 1981, 1992, McQueen et al. 1986, 1989, Ginzberg and Akçakaya 1992, Hunter and Price 1992, Power 1992, Carpenter and Kitchell 1993) or assume that all species at a given trophic level are similarly affected by these changes (e.g., Hairston et al. 1960, Menge and Sutherland 1976, 1987). Although these models have made tremendous contributions to stimulating thought and empirical research, they tend to blur important differences between species within a given trophic level. Natural communities have many species at similar trophic positions (Winemiller 1990, Polis 1991), and each may respond quite differently to the same perturbation (e.g., Berquist and Carpenter 1986, Vanni 1987, Leibold 1989, 1991, Wootton 1992, 1994). Also, comparisons of assemblages in different communities indicate that major environmental changes can lead to the extirpation and replacement of some species but leave other species at the same trophic level largely unaffected (see papers cited in the first paragraph of the Introduction, above). Expanding current theoretical constructs to address differences in species responses within a given trophic level will provide much greater power for predicting changes in food web interactions associated with major environmental perturbations. In this paper I examine why different damselfly genera inhabiting littoral lake communities respond differently to alterations in food web interactions caused by changing the top predator. Characteristic species assemblages for most taxa exist over different ranges along the gradient of permanence from small, vernal ponds to large, permanent lakes (see recent reviews by Wellborn et al. 1996, Skelly 1997). Damselflies in the family Coenagrionidae (Odonata: Zygoptera) are midtrophic level consumers in the littoral food webs of marshes, ponds, and lakes at the permanent end of this environmental gradient. Coenagrionid damselflies are restricted to relatively permanent water bodies, because they require 10 11 mo to complete the aquatic phase (egg and larva) of their life cycle. One group of species in the genus Enallagma is found as larvae only in fishfree water bodies, whereas the remaining Enallagma species are found as larvae only in water bodies that support fish populations (Johnson and Crowley 1980, McPeek 1989, 1990a). In contrast, larvae of Ischnura species, the sister genus of Enallagma (J. M. Brown and M. A. McPeek, unpublished manuscript), are common in both lake types (Johnson and Crowley 1980, McPeek 1990a). Previous work indicates that the differential vulnerabilities of Enallagma species to fish and dragonfly predators are primarily responsible for their segregation between the two lake types (Pierce et al. 1985, Blois-Heulin et al. 1990, McPeek 1990a, b). Many larger ponds, marshes, and lakes that are relatively permanent in their persistence (drying on a time scale of decades or centuries) lack fish, because these water bodies lack inlet and outlet streams to serve as routes of fish colonization or because fish are excluded by the abiotic conditions of the waters (e.g., low winter oxygen concentrations) (Tonn and Magnuson 1982, Rahel 1984). In these fishless waters, large, active dragonfly species (e.g., Anax, Aeshna, and Tramea species) are the top predators in the littoral food web, and dragonfly predation excludes Enallagma species that are found only coexisting with fish (McPeek 1990a). (Hereafter, I will refer to these permanent ponds and lakes without fish as dragonfly lakes, because of dragonflies pivotal role in setting species composition.) Likewise, fish are the top predators in fish lakes, and fish predation excludes Enallagma species that are found only in dragonfly lakes (McPeek 1990a). Fish predation also causes the assemblage of large, active dragonflies found in dragonfly lakes to be replaced by an assemblage of smaller and less active dragonfly species (e.g., Basiaeschna, Epitheca, and Celithemis species) in fish lakes (Hall et al. 1970, Johnson and Crowley 1980, Crowder and Cooper 1982, Morin 1984a, b, Pierce 1988, McPeek 1990a, Werner and McPeek 1994). Interestingly, a third group of dragonflies (e.g., Erythemis and Pachydiplax species) are, like Ischnura, found in both lake types (Johnson and Crowley 1980, McPeek 1990a). Dragonfly predation in fish lakes, abiotic factors, and competitive interactions between Enallagma species do not contribute to maintaining the segregation of Enallagma species (Pierce et al. 1985, McPeek 1990a). The study presented here was designed to reveal why Enallagma and Ischnura species are differentially affected by changing the top predator in the littoral food web by comparing the types and strengths of interactions affecting their mortality and growth in dragonfly and fish lakes. Enallagma and Ischnura species have similar diets (Pearlstone 1973, Thompson 1978, Johnson et al. 1984), and their growth rates have been shown to decrease as damselfly densities increase (Johnson et al. 1984, Anholt 1990, McPeek 1990a). This suggests that in each lake type Enallagma and Ischnura species face the same predators and may compete for the same resources. Theory predicts that coexistence of such species should be facilitated by trade-offs in their abilities to engage in these interactions; the likelihood of coexistence is enhanced if some species are better adapted to avoiding predators and thereby suffer lower mortality due to predation, but other species are better adapted to utilizing resources and thereby have higher growth or fecundity (Levin 1970, Phillips 1974, Vance 1978, Leibold 1989, 1996, Holt et al. 1994, McPeek 1996a). Therefore, in each lake type one genus should have higher fitness components associated with pred-

February 1998 CHANGING THE TOP PREDATOR 3 ator avoidance, while the other should have higher fitness components associated with better resource utilization abilities. Can theory also suggest which roles Enallagma and Ischnura are expected to fill in each lake type? Substitution of one top predator for another will most likely remove the advantage enjoyed by species that are better adapted to predator avoidance; species occupying this position in the food web with one top predator are expected to be extirpated and replaced by other species that are well adapted to avoiding the new top predator (McPeek 1996a). In contrast, species that are effective at exploiting resources but poor at avoiding various types of predators may often be capable of existing in both communities, if positions for such species are available in the food webs of both communities (McPeek 1996a). Given the distributions of Enallagma and Ischnura between dragonfly and fish lakes, these theoretical considerations predict that in each lake type Enallagma species should experience lower mortality due to predation than Ischnura, but in both lake types Ischnura should utilize resources more effectively to give them higher growth rates than the native Enallagma. To test these predictions, I performed a series of complementary laboratory and field experiments to quantify the interactions affecting Enallagma and Ischnura in the two lake types. Duplicate sets of experiments were performed in the two lake types to quantify the relative strengths of competition within and between the two genera, the degree of resource limitation, and the impacts of predators on the survival and growth of native Enallagma and Ischnura. This type of study has been called a comparative experimental approach (Lubchenco and Real 1991, Menge et al. 1994). This approach has many advantages over other possible study designs (e.g., adding fish to a previously fishless lake): (1) because the rate of change from one community type to the other after adding or removing a top predator will be largely determined by how fast missing species colonize experimental units, many problems with lengthy transient dynamics are alleviated; (2) experiments are done in a natural background of abiotic conditions and species composition for each system; and (3) most importantly the interactions are being compared in welldeveloped, natural systems. MATERIALS AND METHODS Field abundance To quantify the distributions of Enallagma and Ischnura species, I sampled larval damselfly densities in three dragonfly lakes and three fish lakes during September and October 1987. Because abundances of littoral invertebrates vary with the structural complexity of the macrophytes in which they are sampled (Cyr and Downing 1988, Rasmussen 1993, Lalonde and Downing 1992), I sampled only lakes dominated by the macrophyte Chara vulgaris in order to standardize density estimates to a common level of structural complexity. I used a 20 cm diameter section of stovepipe covered at one end by two layers of mosquito netting (0.6 1.2 mm mesh) (McPeek 1990a). Samples were taken by plunging the open end of the stovepipe through the Chara into the sediment, sealing the bottom end with a plexiglass plate, and placing the entire contents in a bucket. Samples were returned to the laboratory and sieved through 5.7 mm and 0.5 mm mesh sieves to remove large plant material and sediment. All damselflies were then picked alive from the samples. Five to six samples were taken in each lake at positions chosen haphazardly in water 0.25 to 0.75 m deep. Laboratory feeding bias I evaluated the feeding biases of fish and dragonfly predators characteristic of each lake type in the laboratory: Aeshna mutata dragonflies from dragonfly lakes, and bluegill sunfish (Lepomis macrochirus) and Epitheca cynosura dragonflies from fish lakes. Aeshna is a common dragonfly that is restricted to dragonfly lakes, and Epitheca is most common in fish lakes (Johnson and Crowley 1980, McPeek 1990a). Bluegill sunfish are the dominant fish foraging in the littoral zones of lakes in southwestern Michigan, where this study was conducted (Brown and Ball 1942, Cooper et al. 1971, Werner et al. 1977, Werner and Hall 1988). All three of these predators display feeding biases similar to those of other fish and dragonflies, respectively (McPeek 1990a). The methods used here closely followed those of McPeek (1990a). Feeding trials involving dragonflies were performed in 20 cm diameter circular dishes filled to a depth of 6 cm with tap water. The dishes were bare except for a layer of fiberglass window screening covering the bottoms to provide footing for the odonates. (Trials in which the macrophyte Chara vulgaris was added to dishes to provide structural complexity gave similar results [McPeek 1990a].) For each of 12 trials involving Aeshna mutata, five Enallagma boreale, and five Ischnura verticalis larvae taken from a dragonfly lake, were added to a dish and allowed to acclimate for 3 5 h. One final-instar Aeshna larva was then added to each dish and allowed to feed until it had eaten roughly half the larvae or until satiated, whichever came first; most trials lasted 30 45 min. For each of eight trials involving Epitheca cynosura, five Enallagma signatum, and five Ischnura verticalis larvae taken from a fish lake, were added to a dish and allowed to acclimate for 3 5 h. Then one final-instar Epitheca larva was added to each dish and allowed to feed for 2 d (Epitheca are much less active than Aeshna and consequently feed at much slower rates). No larvae were killed during the acclimation periods when dragonflies were absent, indicating that death due to damselfly predation was minimal during these trials. No dragonfly or damselfly was used in more than one trial.

ECOLOGICAL MONOGRAPHS Saturday Oct 10 01:01 PM Allen Press DTPro emon 68 103 Mp 4 File # 03sc 4 MARK A. McPEEK Ecological Monographs Vol. 68, No. 1 Trials using bluegill sunfish (75 85 mm Standard Length [SL]) were performed in 38-L aquaria. A thin layer of the macrophyte Chara vulgaris was added to each aquarium to provide some structure. For each of nine trials, eight Enallagma signatum and eight I. verticalis larvae from a fish lake were added to an aquarium and allowed to acclimate for 3 5 h. One fish was then added to each aquarium, allowed to feed for 24 h and removed, and the contents of the aquarium was sorted to recover all uneaten damselflies. No fish or damselfly was used more than once. Predator feeding biases were evaluated using Manly s index of preference calculated for Ischnura larvae (Manly 1974, Chesson 1983). Manly s index ( I ) ranges in value from to 1.0. A value of I 0.5 indicates the predator shows no bias towards either prey; I 0.5 indicates that the predator feeds disproportionately on Ischnura larvae. Field experiments General methods. In 1993, the dragonfly-lake field experiments were performed in Pond 3 (a fishless pond of area 1 ha) on the Lux Arbor Reserve of the Kellogg Biological Station (KBS), Barry County, Michigan, USA. Because of a severe blue-green algal bloom in this pond in the summer of 1994, the remaining set of dragonfly-lake experiments were performed in 1995 in another fishless pond on the Lux Arbor reserve, Gravelpit Pond, of area 0.25 ha. All field experiments requiring a lake with fish were performed in Palmatier Lake (Barry County, Michigan, USA), an 6 ha lake that contains a diverse fish assemblage. Bluegill sunfish are the dominant fish preying on littoral zone invertebrates in this lake (Osenberg et al. 1988, Mittelbach and Osenberg 1993), and Chara vulgaris was the dominant submerged macrophyte growing in the littoral zone (M. A. McPeek, personal observation). The methods used in these experiments closely followed those used in previous studies (e.g., McPeek 1990a). Experiments involving no predators or dragonfly predators were done in smaller cylindrical cages (30 cm diameter, 1.2 m high, bottom area 73 m 2 ); experiments involving fish were done in larger cylindrical cages (54 cm diameter, 1.2 m high, bottom area 0.224 m 2 ). All cages were cylinders of 2 cm mesh chicken wire covered with mosquito netting (0.6 1.2 mm mesh size) and sealed at the bottom ends with plastic dishes containing 2 cm of sediment. The tops of cages extended out of the water and were uncovered. All cages in an experiment were linearly arranged in 0.6 0.8 m deep water. Chara vulgaris was added to each cage in natural density. In the absence of predators, damselflies have similar growth and mortality rates in the two cage sizes, and growth and mortality rates in the presence of predators are quite similar to rates in natural populations (McPeek 1990a). All experiments were initiated in late August after damselfly species used in the experiments were too large to pass through the mosquito netting, except in three cases. This procedure minimizes contamination of cages by damselflies (McPeek 1990a). To initiate an experiment, all cages were installed in the lake, Chara was added, and then the cages were allowed to stand for 1 week prior to the addition of damselflies or the application of treatments. This allows colonization by small organisms, which are prey for damselflies and their predators, through the mosquito netting. Chironomids, littoral cladocerans, littoral copepods, ephemeropterans, annelids, and amphipods were all abundant at the end of experiments. Replicates of treatments were always randomly assigned to cages. Experiments were generally terminated in early to mid October. Damselfly larvae used in experiments in the dragonfly lake were collected from fishless water bodies near KBS. Larvae were collected from ponds other than those used for experiments, because collection damages macrophyte beds in the vicinity of the collection, and I was concerned that this disruption in a small pond might influence the results of experiments. Enallagma larvae for experiments in the dragonfly lake were collected from Marshfield Road Marsh, a large fishless marsh just north of KBS. These larvae were a mixture of E. boreale and E. cyathigerum; these two species will be treated as a single taxon in this study, because they are nearly indistinguishable as larvae, they cooccur in all dragonfly lakes used in this study, and they are also very similar in behavior and morphology (McPeek 1989 and unpublished data). Hereafter, I will refer to them together as E. boreale, because E. boreale constituted 90% of the larvae at Marshfield Road Marsh where these species were collected for experiments (M. A. McPeek, unpublished data). I. verticalis larvae were collected from a number of fishless ponds at the Experimental Pond Laboratory of KBS. (I. verticalis is also difficult to distinguish from the closely related I. posita as very small larvae. Nearly all larvae recovered from cages in the field experiments ( 98%) were I. verticalis.) Damselfly larvae used in experiments in the fish lake were collected from Palmatier Lake, but in areas distant from where experiments were established. Two to three Enallagma species were usually included in experiments in fish lakes to mimic the diverse Enallagma assemblage found in fish lakes in North America (e.g., 11 Enallagma species coexist in Palmatier Lake, with three species predominating), but different combinations of species were used in different experiments. This was necessitated by the availability of various species at the times when experiments were established. I. verticalis larvae were used in all experiments. Larvae were assigned randomly by species to the different replicates of an experiment. Larvae were added in the natural size distribution available at the start of the experiment; this meant that for most species, larvae in a narrow range of 2 3 instars were added

February 1998 CHANGING THE TOP PREDATOR 5 (damselflies have 11 instars total). A sample of 20 30 larvae of each species was preserved in 10% neutral formalin to characterize the initial size distribution added to each experiment. At the end of an experiment, the contents of cages (except sediment) were returned to the laboratory where they were sieved through sieves of 5.7 mm and 0.5 mm mesh to remove large plant material and residual sediment. All damselflies, dragonflies, and fish were immediately removed from the sieved samples alive and then preserved in 10% neutral formalin. The head widths of all recovered larvae were measured with a dissecting microscope fitted with an ocular micrometer. All larvae were then dried in a 60 C drying oven for 24 h and individually weighed to determine their dry body mass. When preserved in formalin, damselflies often lose their caudal lamellae. Therefore, the caudal lamellae of all larvae were removed before drying. When the experimental design permitted, multivariate analyses of variance (MANOVA) with a priori, orthogonal contrasts were performed on the mortality and growth rates of all species in cages using the GLM procedure of SAS (SAS 1990). A mortality rate was calculated for each species in each cage in an experiment using mortality rate (ln[number recovered] ln[initial number])/(duration of experiment), where ln signifies natural logarithm. This equation assumes a constant mortality rate throughout the experiment (i.e., N(t) N(0)exp( dt), where d mortality rate, N(0) is the number of larvae added to a cage, N(t) is the number of larvae recovered, and t is the duration of the experiment in days), and this mortality rate (larval deaths per larva per day) is expressed in units of d 1. A growth rate was also calculated for each species in each cage in the experiment. To calculate growth rate, the dry masses of recovered larvae were naturallog transformed, and a mean ln of larval dry mass was calculated for the cage. The growth rate was calculated by growth rate ([mean ln M of recovered larvae] [mean ln M of larvae in initial sample])/(duration of experiment in days), and is expressed in units of d 1, where M is dry mass. This metric of growth rate assumes a model of M(t) M(0)exp(gt), where g is the growth rate and is independent of the initial sizes of species. The F-approximation of Wilks lambda is reported for results of MANOVA. When necessary, based on the results of the MANOVA, variables for each species were analyzed separately using univariate analyses of variance to determine which variables and species contributed to the overall MANOVA treatments effects. All statistical tests for field experiments are two-tailed. Four experiments were performed in the dragonflylake system, and six experiments were performed in the fish-lake system. Two experiments in each lake type quantified the strengths of competitive interactions among the damselflies in the absence of predators and Design of the Intra- vs. Intergeneric Competition Experiment in a study of the effects of changes in the upper levels of a lake food web. An is placed in each of the nine density combinations included in this experiment. TABLE 1. Number of Ischnura added 0 15 30 45 Number of Enallagma added 0 15 30 45 the degree of food limitation. Two experiments were also performed in each lake type to quantify interaction strengths among damselflies and native dragonfly predators, and the degree to which mortality and growth of damselflies in the presence of dragonflies depended on damselfly density. Two experiments also addressed these issues using fish predators in the fish-lake system. In the following I describe the rationale and designs of these duplicate experiments. Intrageneric vs. intergeneric competition. These experiments were designed to quantify competition among Enallagma and Ischnura species in the absence of predators. The same basic design was duplicated in each lake type in 1993. Twenty-seven cages were installed in a lake, and following the 1-wk prey colonization period damselfly larvae were added to cages in the density combinations given in Table 1. In this experimental design, three total damselfly abundances were established (i.e., 15, 30, and 45 larvae/cage), and the relative abundances of Enallagma and Ischnura were manipulated within each total abundance. This experiment was designed to quantify competitive interactions between the two genera; it was not designed to partition competitive effects among species within each genus. Three replicates of each density combination were performed. The natural density for this water volume at this time of year would be equivalent to 50 larvae/cage. These densities were chosen because previous experiments have shown that competitive effects on growth are detected at densities well below natural (generally 25 larvae/cage), but growth rates do not continue to decline as density is increased above natural densities (McPeek 1990a). In the experiment done in the dragonfly lake Enallagma boreale and I. verticalis were used, and the experiment was terminated after 43 d. In the experiment done in the fish lake, E. geminatum and E. vesperum comprised the Enallagma added to cages: for the 15 Enallagma treatments, eight E. vesperum and seven E. geminatum were added; for the 30 Enallagma treatments, 15 of each were added; and for the 45 Enallagma treatments, 23 E. vesperum and 22 E. geminatum were added. The fish-lake experiment was terminated after 40 d. Because larvae of one or the other species were missing from some treatments, an overall MANOVA could

ECOLOGICAL MONOGRAPHS Saturday Oct 10 01:01 PM Allen Press DTPro emon 68 103 Mp 6 File # 03sc 6 MARK A. McPEEK Ecological Monographs Vol. 68, No. 1 not be performed on this design. Separate MANOVAs were therefore performed on growth and mortality rates for each species in each experiment. Density manipulation in the presence of dragonflies. These experiments test whether dragonflies impose density-dependent mortality on the damselflies, and whether damselflies compete in the presence of dragonflies. The same basic experimental design was duplicated in each lake type in 1993. Nine 30 cm diameter cages were installed in a lake, and after the prey colonization period, three total damselfly density treatments of 15, 45, or 90 total damselflies/cage were established, with three replicates/treatment. These densities bracketed natural density ( 50 larvae/cage). The relative abundances of species initially present in cages were held constant across the density treatments. In the dragonfly-lake experiment, the relative abundances of the two species added to cages were held constant across the total density treatments at a ratio of 2:1 Enallagma boreale : Ischnura verticalis. One penultimate instar Anax junius dragonfly larva (Anisoptera: Aeshnidae) was added to each cage 8 d after the damselflies were added, and the experiment was terminated after 35 d. In the fish-lake experiment, equal numbers of I. verticalis, Enallagma vesperum and E. geminatum were initially present in each cage. One penultimate instar Basiaeschna janata dragonfly larva (Anisoptera: Aeshnidae) was added to each cage 3 d after the damselflies were added, and the experiment was terminated after 30 d. Dragonfly densities are near natural for these species in their respective lakes (McPeek 1990a). Density manipulation in the presence of fish. This experiment is analogous to the previous pair involving dragonflies, but tests whether fish impose density-dependent mortality on the coexisting damselflies, and whether the coexisting damselflies compete in the presence of fish. Six 54 cm diameter cages were installed in the fish lake in 1995, and following the 1-wk prey colonization period, two total damselfly density treatments of 45 and 225 total damselflies/cage were established, with three replicates/treatment. These densities bracketed the natural density for the cage size used ( 150 larvae/cage). In each treatment, Enallagma vesperum initially comprised 55.6% of the larvae, E. geminatum initially comprised 22.2%, and I. verticalis initially comprised 22.2%. One bluegill sunfish (Lepomis macrochirus) (75 80 mm SL) was added to each cage 3 d after the damselflies were added; this fish density is within the natural range for lakes in southwestern Michigan (Mittelbach 1988). This experiment was terminated after 44 d. Food addition. Two experiments were performed in 1995 to evaluate whether damselfly mortality and growth rates are limited by food availability in either lake type. In the dragonfly lake, two total damselfly density treatment levels ([1] 15 Enallagma boreale and 7 I. verticalis or [2] 45 E. boreale and 21 I. verticalis) were cross-factored with two levels of food addition ([1] no food added, [2] food added). Four replicates per treatment combination were performed. This experiment was terminated after 51 d. Results of the Density Manipulation in the Presence of Dragonflies experiment done in the fish lake indicated that competitive interactions among damselflies are more pronounced in the presence of dragonflies (see Results). Therefore, the experiment done in the fish lake to test for food limitation was designed to incorporate both damselfly density and dragonfly predator effects. Three damselfly abundance/dragonfly treatments ([1] low damselfly density and dragonflies absent, [2] high damselfly density and dragonflies absent, and [3] high damselfly density and dragonflies present) were crossfactored with two levels of food addition ([1] no food added, [2] food added) in this experiment. Three replicates per treatment were performed. Low damselfly density treatments initially had 20 damselfly larvae present; high damselfly density treatments initially had 80 damselflies present (natural density 50 larvae/ cage). In this experiment, Enallagma geminatum and E. vesperum were used. In all cages I. verticalis initially comprised 26.7% of damselflies, E. geminatum 21.3%, and E. vesperum 52.0%. Four penultimate instar Epitheca cynosura larvae (Anisoptera: Libellulidae) were added to each cage in the High Damselfly Density and Dragonflies Present treatment combinations 3 d after the damselflies were added. Epitheca larvae were used because Basiaeschna larvae were not available. Also, four were included because of Epitheca s smaller size. This density of dragonflies is near natural for total dragonfly abundances in fish lakes (McPeek 1990a). This experiment was terminated after 45 d. Food additions in both of these experiments were accomplished by adding large quantities of limnetic zooplankton to cages. Zooplankton were collected from the limnetic zone of Lawrence Lake (Kalamazoo County, Michigan, USA) on mornings when food was to be added. I chose to use limnetic zooplankton from Lawrence Lake for food additions for three reasons: (1) damselfly larvae readily eat limnetic zooplankton species in the laboratory; (2) most of these prey individuals were too large to pass through the mosquito netting; and (3) large quantities of limnetic zooplankton could be collected quickly. All zooplankton collected in three vertical net tows over the entire water column ( 12 m depth) with a zooplankton net (30 cm diameter opening, 125- m mesh size) were added to each food addition cage in both experiments. An equivalent amount of Lawrence Lake water, which had been strained through the zooplankton net to remove zooplankton, was added to each cage that had no food addition as a control for any disturbance or water chemistry effects. Food was added weekly (7, 14, 21, and 28 September, 5 and 12 October) for the duration of both experiments. Interactions with dragonflies. These experiments

February 1998 CHANGING THE TOP PREDATOR 7 quantified mortality inflicted by coexisting dragonflies and nonlethal effects of the presence of dragonflies (e.g., feeding interference or exploitative competition between the damselflies and their dragonfly predators) on the damselflies. In the dragonfly lake, 40 Enallagma boreale and 15 I. verticalis were added to each of the 14 cages in 1995. Three treatments ([1] No Dragonfly, [2] Caged Dragonfly, or [3] Free-Ranging Dragonfly) were then established in these cages. Four replicates each were performed for the No Dragonfly and the Caged Dragonfly treatments, and six replicates were performed for the Free-Ranging Dragonfly treatment. For the Caged Dragonfly treatment, one antepenultimate-instar Anax junius larva was placed inside a small enclosure, and the enclosure was placed into the cage. The dragonfly enclosure was 11 11 6 cm, and was constructed by placing a small, coarse-mesh (openings 1.7 1.0 cm), plastic produce container inside a bag constructed of mosquito netting. A glass dowel was placed inside the enclosure for a perch. This enclosure allowed damselflies to detect that a large dragonfly was present by both vision and olfaction, but prevented the dragonfly from eating the damselflies or the damselflies prey. Identical enclosures without a dragonfly were placed in all other cages. For the Free-Ranging Dragonfly treatment, one unrestrained, antepenultimate-instar Anax was placed in the cage. No dragonflies were added to the No Dragonfly treatment cages. This experiment was terminated after 46 d. The cages in this experiment for some unknown reason were significantly contaminated by Ischnura moving through the mosquito netting. Consequently, the results for Ischnura in this experiment are unreliable and will not be presented. In the fish lake in 1994, 15 Ischnura verticalis, 15 Enallagma geminatum, 15 E. vesperum, and 5 E. hageni were added to each of 14 cages. Four days later one of the three predator treatments described above was assigned to each cage, except that penultimate-instar Basiaeschna janata larvae were used instead of Anax junius. This experiment was terminated after 46 d. If dragonflies compete exploitatively with damselflies, damselfly growth should be decreased in the Free- Ranging Predator treatment as compared to the other treatments, because this is the only treatment in which a dragonfly is free to consume prey in the cage. If the presence of dragonflies generates feeding interference in the damselflies (i.e., reduced feeding in the presence of a predator), damselfly growth should also be lower in the Caged Dragonfly treatment as compared to the No Dragonfly treatments. If both exploitation and interference are important, treatment effects should be most pronounced in the Free-Ranging Dragonfly treatment where both exploitation and interference can operate, but treatment effects due to interference should still be detected in the Caged Dragonfly treatment. Interactions with fish. This experiment is analogous to the Interactions with Dragonflies experiments, but tests for comparable effects due to fish in the fish lake. In 1994, the same number of damselflies were added to each of 14 cages as follows: 30 Ischnura verticalis, 40 Enallagma geminatum, 40 E. vesperum, and 15 E. hageni. This total damselfly density of 125 larvae/cage is somewhat less than the natural density for the cages used ( 150 larvae/cage). Five days later one of three predator treatments was constructed in each cage. The three treatments were [1] No Bluegill, [2] Caged Bluegill, or [3] Free-Ranging Bluegill. Four replicates each were performed for the No Bluegill and the Caged Bluegill treatments, and six replicates were performed for the Free-Ranging Bluegill treatment. For the Caged Bluegill treatment, one bluegill (50 65 mm SL) was placed inside a small enclosure, and the enclosure was placed into the cage. The predator enclosures used in this experiment were 25 25 25 cm, and were constructed of mosquito netting bags around 12 mm diameter PVC (polyvinyl chloride plastic) pipe frames. This again prevented the bluegill from feeding on the damselflies or the damselflies prey in the cage. Empty containers were placed in cages of the other two treatments. One bluegill of similar size was placed in each of the Free-Ranging Bluegill treatment cages, and no bluegills were added to the No Bluegill treatment cages. This experiment was terminated after 41 d. RESULTS Field abundances The quantitative density estimates obtained in this study are consistent with patterns observed in previous studies (Johnson and Crowley 1980, McPeek 1990a). Larval Enallagma aspersum, E. boreale, and E. cyathigerum were found only in the three dragonfly lakes (Table 2). Five other Enallagma species were found only in the three fish lakes (Table 2). A total of thirteen Enallagma species inhabit fish lakes of southwestern Michigan (M. A. McPeek, unpublished data), but only these five were common enough to be detected by the sampling methods. Enallagma species constituted 85% of the total damselfly assemblage in all six lakes. Ischnura species were abundant in both lake types (Table 2), and Ischnura abundances in the two lake types were similar (t 4 0.89, P 0.40 for ln mean abundances in the two lake types). Laboratory feeding bias The dragonflies from both lake types and bluegill sunfish all imposed substantially greater mortality on Ischnura larvae than on Enallagma species with which each predator naturally coexists (Table 3). On average, 65 70% of larvae consumed by both dragonflies were Ischnura larvae, and 80% of damselflies consumed by bluegill sunfish were Ischnura larvae (Table 3). Ischnura larvae were more vulnerable to both dragonflies and fish than were Enallagma species that naturally coexist with these predators.

ECOLOGICAL MONOGRAPHS Saturday Oct 10 01:01 PM Allen Press DTPro emon 68 103 Mp 8 File # 03sc 8 MARK A. McPEEK Ecological Monographs Vol. 68, No. 1 Larval damselfly density estimates (number of larvae/m 2 lake bottom) from three dragonfly lakes and three fish lakes in southwestern Michigan in a study of the effects of changes in the upper levels of a food web. TABLE 2. Dragonfly lakes Fish lakes Marshfield Three Lakes Damselfly species Turkey Marsh Marsh Reservoir II Palmatier Hamilton Enallagma aspersum 12.3 27.6 6043.9 1160.9 E. boreale/ cyathigerum 128.5 59.8 166.5 131.9 87.4 36.0 E. antennatum 82.2 57.4 5.1 12.6 10.3 15.9 E. geminatum 82.2 25.2 539.6 329.1 426.6 249.3 E. hageni 195.3 72.1 15.4 25.8 159.3 100.2 E. signatum 215.9 115.4 308.4 428.2 138.8 77.4 E. vesperum 20.6 25.2 66.8 53.1 10.3 25.2 Ischnura posita/ verticalis 15.4 25.8 30.8 37.8 200.4 134.8 5.1 12.6 30.8 47.8 41.1 25.2 Number of samples 6 5 6 6 6 6 Notes: Data are means 1 SD. The samples were collected between 26 September 1987 and 9 October 1987. Field experiments Intrageneric vs. intergeneric competition. In these experiments done in both lake types in the absence of predators, significant treatment effects consequent to manipulations in total damselfly density were apparent (some a priori, orthogonal contrasts in the MANOVA of each experiment with P 5 for each genus), but these responses were due primarily to differences in growth and not mortality. Manipulations of total damselfly density and generic relative frequency had no consistent effects on the mortality rates of any species in either experiment (all contrasts P 5). In the dragonfly-lake experiment, Ischnura growth rates overall were 655% higher than those of Enallagma boreale (Fig. 1A, B). Compared to the low total damselfly density, Ischnura growth rates were reduced by 12% in the medium total density treatments (contrast testing low vs. medium and high total density treatments: F 1,12 9.47, P 1) and by 24% in the high total density treatments, although this latter comparison was not statistically significant (contrast testing medium vs. high total density treatments, F 1,12 3.35, P 9; Fig. 1A). Ischnura growth rates did not differ among relative frequency treatments within the medium or high total densities (all contrasts P 0.25; Fig. 1A). E. boreale larvae displayed a significant reduction in growth rates only at the high total density treatments (50% reduction in growth rate from the low to high total density treatments; contrast testing medium vs. high total density treatments, F 1,12 9.69, P 1; Fig. 1B). As with Ischnura, relative frequency had no effect on E. boreale growth rates when total density was constant (all contrasts P 5; Fig. 1B). The results of the comparable experiment done in the fish lake were similar to those of the dragonfly-lake experiment. In the fish-lake experiment, Ischnura growth rates overall were 160% higher than those of E. geminatum and 680% higher than E. vesperum (Fig. 1C, D). As compared to the low total damselfly density, Ischnura growth rates were 20% lower in the high total density treatments, but this difference was only marginally significant (contrast testing medium vs. high Results of predator feeding bias experiments for dragonflies and fish from dragonfly and fish lakes. TABLE 3. Predator Enallagma species Prey Ischnura species N I Dragonfly lakes Aeshna mutata E. boreale I. verticalis 12 0.64 5* Fish lakes Epitheca cynosura Lepomis macrochirus E. signatum E. signatum I. verticalis I. verticalis 8 9 0.69 6* 0.84 0.10* * P 5 for t test of I 0.50. N number of replicates of that particular set of species. Manly s index of preference ( I ) for the predator feeding on Ischnura larvae in the trials. Manly s index ranges in value from 0 to 1.00; in this experiment I is equivalent to the proportion of consumed damselflies that were Ischnura. A value of I 0.50 in these experiments indicates that the predator displayed no bias in feeding on the two damselfly species. A value of I 0.50 indicates that the predator imposed significantly greater mortality on Ischnura larvae than on Enallagma larvae in the trials. Means 1 SE are given with significance noted.

February 1998 CHANGING THE TOP PREDATOR 9 FIG. 1. Growth rates of the damselflies included in the Intrageneric vs. Intergeneric Competition experiments (which were designed to quantify competition among Enallagma and Ischnura species in the absence of predators) in both lakes. For comparison, the results of the experiment done in the dragonfly lake are given in the left column of panels, and the results of the fish-lake experiment are given in the right-hand column of panels. To permit comparisons between experiments I have given damselfly densities expressed in units of larvae/m 2 of lake bottom. Symbols representing the same total number of damselflies of both genera initially added to cages are connected by lines in the graphs and are given above the panels. The natural density of larvae at the beginning of the experiments was 685 damselflies/m 2 of lake bottom. The abscissa identifies the relative frequencies of Ischnura and Enallagma larvae added to cages. Symbols are slightly offset from one another to aid in interpretation. Each species is only present in six of the nine treatments (Table 1). total density treatments, F 1,12 4.33, P 6; Fig. 1C), but Ischnura growth rates did not differ among the various treatments at the low or medium total densities (all contrasts P 0.20) or among the relative frequency treatments within the high total density (all contrasts P 0.85). E. vesperum growth rates were unaffected by all density and relative frequency treatments (all contrasts P 0.60; Fig 1C). E. geminatum larvae had a 160% higher growth rate when at low density in the absence of Ischnura as compared to the other treatments (contrast testing low vs. medium and high total density treatments, F 1,12 16.86, P 01), but E. geminatum growth rates did not differ among any of the other treatments (other contrasts P 0.20; Fig. 1D). In both lake types, damselflies clearly compete with one another, because increasing total damselfly density causes decreases in their growth rates. However, Ischnura and Enallagma species appear to be relatively equivalent competitors, since the relative abundances of the two genera do not affect growth rates. Density manipulation in the presence of dragonflies. In the experiment done in the dragonfly lake, Anax dragonfly larvae were not recovered from two cages in the high total damselfly density treatment; these individuals apparently died sometime during the experiment. These two cages were therefore excluded from analyses. In the dragonfly-lake experiment, I was unable to perform an overall MANOVA because the loss of the two cages reduced the total degrees of freedom. However, univariate analyses indicate that increasing total damselfly densities had no effect on Enallagma boreale or Ischnura mortality rates (overall treatment effects for both species P 0.30; Fig. 2A) or on their growth rates (overall treatment effects for both species P

ECOLOGICAL MONOGRAPHS Saturday Oct 10 01:01 PM Allen Press DTPro emon 68 103 Mp 10 File # 03sc 10 MARK A. McPEEK Ecological Monographs Vol. 68, No. 1 FIG. 2. Rates of mortality and growth for the damselfly species included in the Density Manipulation in the Presence of Dragonflies experiments done in each lake. The species of dragonfly used in each experiment is given above the column of panels. The abscissa gives the total number of damselflies initially added to cages. The natural density of larvae at the beginning of the experiments was 685 damselflies/m 2 of lake bottom. Symbols are the means 1 SE for each species in a treatment. Symbols are slightly offset from one another to aid in interpretation. 0.10; Fig. 2B). I also calculated Manly s preference index ( I ) for the number of larvae of each species recovered from each cage to test explicitly for a difference in mortality rate between the two genera. Averaged across all density treatments, Manly s index for Ischnura was I 0.62 8 (mean 1 SE), which is not statistically significantly different from no preference (t 6 1.47, P 0.10) but is, however, nearly identical to the value measured in the laboratory feeding bias experiment with Aeshna mutata (Table 3). In the experiment done in the fish lake, the contrast testing the low vs. medium and high total damselfly density treatments in the MANOVA was not statistically significant (F 6,1 16.19, P 95). This overall test was not statistically significant because it includes both mortality and growth rates for all three species as response variables. Univariate ANOVAs indicate that mortality rates of all three species were not different among the three total damselfly density treatments (all contrasts P 0.30; Fig. 2C). However, all three species had depressed growth rates at the medium and high total damselfly density as compared to the low density (Fig. 2D): Ischnura growth rates were depressed by 30% (F 1,6 6.94, P 5), E. vesperum growth rates by 70% (F 1,6 14.39, P 1), and E. geminatum growth rates by 35% (F 1,6 6.78, P 5). Growth rates did not differ between the medium and high total density treatments for any species (all P 0.40). Dragonflies do not impose density-dependent mortality on coexisting damselflies in either lake type. Also, competitive effects among damselflies in the presence of foraging dragonflies are only apparent in fish lakes. Density manipulation in the presence of fish. In 1995 an unusually late hatch of E. geminatum caused substantial contamination by many small larvae of this species into cages of this experiment. Therefore, results for E. geminatum will not be reported. In this experiment done in the fish lake, the mortality and growth of both Ischnura and E. vesperum changed with damselfly density in the presence of bluegill sunfish. The design of the experiment did not permit an overall MANOVA test including the mortality and growth rates of both species simultaneously, but separate MANOVAs for each species were done. The combined mortality and growth rate responses of Ischnura